| Message ID | 1557758186-18706-3-git-send-email-stanley.chu@mediatek.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | scsi: ufs: add error handlings of auto-hibern8 | expand |
Hi, Stanley >+ >+static inline bool ufshcd_is_auto_hibern8_error(struct ufs_hba *hba, >+ u32 intr_mask) >+{ >+ return (ufshcd_is_auto_hibern8_supported(hba) && >+ !hba->uic_async_done && Here check if uic_async_done is NULL, no big problem so far, but not safe enough. How about setting a flag in ufshcd_auto_hibern8_enable(), I concern about how to compatible with auto_hibern8 disabled condition. //Bean
Hi Bean, Thanks so much for review. On Mon, 2019-05-13 at 18:21 +0000, Bean Huo (beanhuo) wrote: > Hi, Stanley > > >+ > >+static inline bool ufshcd_is_auto_hibern8_error(struct ufs_hba *hba, > >+ u32 intr_mask) > >+{ > >+ return (ufshcd_is_auto_hibern8_supported(hba) && > >+ !hba->uic_async_done && > > Here check if uic_async_done is NULL, no big problem so far, but not safe enough. > How about setting a flag in ufshcd_auto_hibern8_enable(), > > I concern about how to compatible with auto_hibern8 disabled condition. Currently auto-hibern8 disabling method is not implemented in mainstream, so an "enabling" flag may looks redundant unless disabling path is really existed. I agree that checking hba->uic_async_done here does not look so intuitive. However even if auto-hibern8 is disabled, these checks could be safe enough because both "UIC_HIBERNATE_ENTER" and "UIC_HIBERNATE_EXIT" are raised only if "manual-hibernate" is performed, and in this case hba->uic_async_done shall be true. Anything else or corner case I missed? > > > //Bean Thanks, Stanley > > _______________________________________________ > Linux-mediatek mailing list > Linux-mediatek@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-mediatek
Hi, Stanley Thanks for reply. > >On Mon, 2019-05-13 at 18:21 +0000, Bean Huo (beanhuo) wrote: >> Hi, Stanley >> >> >+ >> >+static inline bool ufshcd_is_auto_hibern8_error(struct ufs_hba *hba, >> >+ u32 intr_mask) >> >+{ >> >+ return (ufshcd_is_auto_hibern8_supported(hba) && >> >+ !hba->uic_async_done && >> >> Here check if uic_async_done is NULL, no big problem so far, but not safe >enough. >> How about setting a flag in ufshcd_auto_hibern8_enable(), > >> >> I concern about how to compatible with auto_hibern8 disabled condition. > >Currently auto-hibern8 disabling method is not implemented in mainstream, >so an "enabling" flag may looks redundant unless disabling path is really >existed. > Did you try to update Auto-Hibernate Idle Timer with 0 through '/sys' (scsi: ufs: Add support for Auto-Hibernate Idle Timer)? I don't know if this will disable your UFS controller Auto-Hibernate. If having a look at UFS host Spec, software writes “0” to disable Auto-Hibernate Idle Timer. Sorry I cannot verify this on my platform since it doesn't support auto-hibernate. >I agree that checking hba->uic_async_done here does not look so intuitive. >However even if auto-hibern8 is disabled, these checks could be safe enough >because both "UIC_HIBERNATE_ENTER" and "UIC_HIBERNATE_EXIT" are >raised only if "manual-hibernate" is performed, and in this case hba- >>uic_async_done shall be true. > Yes, most of cases ,this is no problem. >Anything else or corner case I missed? > The others are fine. I only concern checking hba->uic_async_done. //Bean
Hi Bean, On Tue, 2019-05-14 at 11:14 +0000, Bean Huo (beanhuo) wrote: > Hi, Stanley > Thanks for reply. > > > > >On Mon, 2019-05-13 at 18:21 +0000, Bean Huo (beanhuo) wrote: > >> Hi, Stanley > >> > >> >+ > >> >+static inline bool ufshcd_is_auto_hibern8_error(struct ufs_hba *hba, > >> >+ u32 intr_mask) > >> >+{ > >> >+ return (ufshcd_is_auto_hibern8_supported(hba) && > >> >+ !hba->uic_async_done && > >> > >> Here check if uic_async_done is NULL, no big problem so far, but not safe > >enough. > >> How about setting a flag in ufshcd_auto_hibern8_enable(), > > > >> > >> I concern about how to compatible with auto_hibern8 disabled condition. > > > >Currently auto-hibern8 disabling method is not implemented in mainstream, > >so an "enabling" flag may looks redundant unless disabling path is really > >existed. > > > Did you try to update Auto-Hibernate Idle Timer with 0 through '/sys' (scsi: ufs: Add support for Auto-Hibernate Idle Timer)? > I don't know if this will disable your UFS controller Auto-Hibernate. > If having a look at UFS host Spec, software writes “0” to disable Auto-Hibernate Idle Timer. > Sorry I cannot verify this on my platform since it doesn't support auto-hibernate. > Sorry I missed this /sys interface for Auto-Hibernate control. Yes, I have tested "Auto-Hibernate disabled" case, in this case, UIC_HIBERNATE_ENTER and UIC_HIBERNATE_EXIT interrupts comes only if Manual-Hibernate is performed and waiting for completion. Both interrupts will not be identified as Auto-Hibernate errors by checking hba->uic_async_done. As for your concerning, I would like to make "Auto-Hibernate error detection" more precise in next version: Use below conditions instead of checking hba->uic_async_done: As-is: static inline bool ufshcd_is_auto_hibern8_error(struct ufs_hba *hba, u32 intr_mask) { return (ufshcd_is_auto_hibern8_supported(hba) && !hba->uic_async_done && (intr_mask & UFSHCD_UIC_AH8_ERROR_MASK)); } To-be: static bool ufshcd_is_auto_hibern8_error(struct ufs_hba *hba, u32 intr_mask) { if (!ufshcd_is_auto_hibern8_supported(hba)) return false; if (!(intr_mask & UFSHCD_UIC_AH8_ERROR_MASK)) return false; if (hba->active_uic_cmd && ((hba->active_uic_cmd->command == UIC_CMD_DME_HIBER_ENTER) || (hba->active_uic_cmd->command == UIC_CMD_DME_HIBER_EXIT))) return false; return true; } What would you think about this change? > > >I agree that checking hba->uic_async_done here does not look so intuitive. > >However even if auto-hibern8 is disabled, these checks could be safe enough > >because both "UIC_HIBERNATE_ENTER" and "UIC_HIBERNATE_EXIT" are > >raised only if "manual-hibernate" is performed, and in this case hba- > >>uic_async_done shall be true. > > > Yes, most of cases ,this is no problem. > > >Anything else or corner case I missed? > > > The others are fine. I only concern checking hba->uic_async_done. > > //Bean Many thanks, Stanley
diff --git a/drivers/scsi/ufs/ufshcd.c b/drivers/scsi/ufs/ufshcd.c index 1665820c22fd..e0e3930abc19 100644 --- a/drivers/scsi/ufs/ufshcd.c +++ b/drivers/scsi/ufs/ufshcd.c @@ -5254,6 +5254,7 @@ static void ufshcd_err_handler(struct work_struct *work) goto skip_err_handling; } if ((hba->saved_err & INT_FATAL_ERRORS) || + ufshcd_is_auto_hibern8_error(hba, hba->saved_err) || ((hba->saved_err & UIC_ERROR) && (hba->saved_uic_err & (UFSHCD_UIC_DL_PA_INIT_ERROR | UFSHCD_UIC_DL_NAC_RECEIVED_ERROR | @@ -5431,6 +5432,15 @@ static void ufshcd_check_errors(struct ufs_hba *hba) queue_eh_work = true; } + if (ufshcd_is_auto_hibern8_error(hba, hba->errors)) { + dev_err(hba->dev, + "%s: Auto Hibern8 %s failed - status: 0x%08x, upmcrs: 0x%08x\n", + __func__, (hba->errors & UIC_HIBERNATE_ENTER) ? + "Enter" : "Exit", + hba->errors, ufshcd_get_upmcrs(hba)); + queue_eh_work = true; + } + if (queue_eh_work) { /* * update the transfer error masks to sticky bits, let's do this @@ -5493,6 +5503,10 @@ static void ufshcd_tmc_handler(struct ufs_hba *hba) static void ufshcd_sl_intr(struct ufs_hba *hba, u32 intr_status) { hba->errors = UFSHCD_ERROR_MASK & intr_status; + + if (ufshcd_is_auto_hibern8_error(hba, intr_status)) + hba->errors |= (UFSHCD_UIC_AH8_ERROR_MASK & intr_status); + if (hba->errors) ufshcd_check_errors(hba); diff --git a/drivers/scsi/ufs/ufshcd.h b/drivers/scsi/ufs/ufshcd.h index ecfa898b9ccc..1bd9c8b61ed2 100644 --- a/drivers/scsi/ufs/ufshcd.h +++ b/drivers/scsi/ufs/ufshcd.h @@ -740,6 +740,19 @@ return true; #endif } +static inline bool ufshcd_is_auto_hibern8_supported(struct ufs_hba *hba) +{ + return (hba->capabilities & MASK_AUTO_HIBERN8_SUPPORT); +} + +static inline bool ufshcd_is_auto_hibern8_error(struct ufs_hba *hba, + u32 intr_mask) +{ + return (ufshcd_is_auto_hibern8_supported(hba) && + !hba->uic_async_done && + (intr_mask & UFSHCD_UIC_AH8_ERROR_MASK)); +} + #define ufshcd_writel(hba, val, reg) \ writel((val), (hba)->mmio_base + (reg)) #define ufshcd_readl(hba, reg) \ diff --git a/drivers/scsi/ufs/ufshci.h b/drivers/scsi/ufs/ufshci.h index 6fa889de5ee5..4bcb205f2077 100644 --- a/drivers/scsi/ufs/ufshci.h +++ b/drivers/scsi/ufs/ufshci.h @@ -148,6 +148,9 @@ enum { UIC_HIBERNATE_EXIT |\ UIC_POWER_MODE) +#define UFSHCD_UIC_AH8_ERROR_MASK (UIC_HIBERNATE_ENTER |\ + UIC_HIBERNATE_EXIT) + #define UFSHCD_UIC_MASK (UIC_COMMAND_COMPL | UFSHCD_UIC_PWR_MASK) #define UFSHCD_ERROR_MASK (UIC_ERROR |\
Currently auto-hibern8 is activated if host supports auto-hibern8 capability. However no error handlings are existed thus this feature is kind of risky. If "Hibernate Enter" or "Hibernate Exit" fail happens during auto-hibern8 flow, the corresponding interrupt "UIC_HIBERNATE_ENTER" or "UIC_HIBERNATE_EXIT" shall be raised according to UFS specification. This patch adds auto-hibern8 error handlings: - Monitor "Hibernate Enter" and "Hibernate Exit" interrupts after auto-hibern8 feature is activated. - If fail happens, trigger error handlings just like "manual-hibernate" fail and use the same flow: Identify errors and schedule UFS error handler in ufshcd_check_errors(), and then do host reset and restore in UFS error handler. Signed-off-by: Stanley Chu <stanley.chu@mediatek.com> --- drivers/scsi/ufs/ufshcd.c | 14 ++++++++++++++ drivers/scsi/ufs/ufshcd.h | 13 +++++++++++++ drivers/scsi/ufs/ufshci.h | 3 +++ 3 files changed, 30 insertions(+)